Discussion Methods Of Incipient Fault Detection And Identification In10kV Underground Cables

With the accelerated process of urbanization, a large quantity of power cables have been used.and the cable network structures are becoming increasingly sophisticated. The underground cable which is less susceptible to environmental impact and has low maintenance is more secure than the overhead line in inclement weather, and more economical in short-distance transmission.However,underground cable is easier to be affected by partial discharge and flashover, and gradually deteriorated to a permanent fault, which is difficult to seek. The detection of incipient fault in cables can identify the defective cable. And this could help us reducing the load of the defective cables, maximizing its remaining life, and replacing the cables timely before the permant fault. There is great significance to protect the stability and security of power supply.On the basis of studying cable insulation aging mechanism and the shape and development process of the incipient fault in cables, a incipient fault simulation model in PSCAD/EMTDC is built, and simulated a single phase-to-earth early fault, and the mathematical model and preliminary parameters of simulation model were elaborated in detail. The affection of different initialparameters to fault resistance was analyzed, and the changes of the peak fault current and RMS current caused by the resistance change were analyzed as well. In comparison with the actual measured fault current of the incipient faults in reference, the effectiveness of the cable incipient fault simulation model is reliable.According to the simulated signals get from the cable incipient fault simulation model in PSCAD/EMTDC, wavelet singularity detection and Bayesian changing point analysis were used to detect the incipient fault current signal. The signal singularity is detected with db4wavelet and complex Morlet wavelet. Then find out the start time and ending time of the incipient fault according to the result of the singularity detection.A Bayesian mathematical model of time series change-point analysis was built. The posteriori probability density function of Bayesian mathematical model is solved by Markov Chain Monte Carlo (MCMC), and the maximum probability point of the reasult is the time when fault happens. These two approachs both can detect theincipient fault effectively. The analysis shows that singularity detection based on complex Morlet wavelet is better than db4wavelet. And the result of Bayesian is not as good as wavelet singularity detection when the fault is not obvious. Experimental results show that the Bayesian approach is more robust than wavelet singularity detection approach. The signal of cable incipient fault which differs from a permanent fault is self-recovery, and belongs to transient disturbance. In order to distinguish it from the others, classification is needed. Based on the distribution network established in PSCAD/EMTDC, the transient signals of capacitor switchingwere simulated, which can be used as the comparison signal to identify the incipient fault. Harmonic analysis was made on incipient fault signal and capacitor switching disturbance signal by fast Fourier transform (FFT), and frequency domain feature vector was established based on the ratio of each harmonic and total harmonic distortion, which was used as the basis for distinguish the two types of transient signals. Binary classifiers based on probabilistic neural network (PNN) and support vector machine (SVM) were used to classify and identity the two signals. The simulation results show that the two classifiers can distinguish between incipient fault signal and capacitor switching disturbance signal, and in the case of small samples, SVM classifier has the higher accuracy.